Built by The Quiet Scribe
Free starter kit

Cut short, captioned clips out of your recorded calls.

A small starter kit that runs on your own laptop. You choose the moments by reading the transcript, and the captions follow the spoken words.

Get the starter How it works

What to watch: the captions. Each word lights up as it is spoken. This is a real 20-second clip cut from a screen recording, and the screen text is small in a vertical frame, so open full screen to read it.

The whole job

Four steps

  1. Transcribe

    Whisper, a speech-to-text program, writes down every word with its time. It runs on your laptop.

    my-call.asr.txt

  2. Pick

    Read the transcript and note where each moment starts and stops. You choose by reading, not by scrubbing.

    plan.json

  3. Cut

    The tool cuts those moments out and joins them. The captions travel with the cut.

    cut.py

  4. Render

    Remotion draws the captions onto the video, one word at a time, and writes the finished clip.

    clip.mp4
Under the hood

Why the captions stay in sync

Captions drift when their timing is estimated. Here it is taken from the transcript. Each word carries its own start and end time, so a caption lights up when its word is spoken, not when a sentence begins.

Cuts move off words. If a cut point falls in the middle of a spoken word, the tool moves it just outside that word and tells you which one it moved. It can only do this for words Whisper heard.

Joins have their own rules. A word belongs to whichever piece holds its midpoint, so a word that straddles a join is not left lit with no end. Every join starts a new caption line, so one line never mixes two moments.

A cut point moves out of a word into the gap Top row: a cut lands inside the word "numbers" and clips it. Bottom row: the cut moves to the gap before the word, so no word is clipped. YOU ASK FOR THIS CUT so the numbers were off lands mid-word: clipped syllable YOU GET THIS CUT so the numbers were off snapped to the gap before the word, and it tells you
Try it

Try it

# once, after unzipping the kit
cd call-clip-starter
pip install -r requirements.txt && (cd remotion && npm install)

# every clip
python3 transcribe.py my-call.mp4
python3 cut.py plan.json
cd remotion && npm run render

Needs Python 3.9+, Node 18+ and ffmpeg. Tested on macOS only so far.

One-page explainer: call-clip-starter-explainer.pdf

Free starter kit

Where should we send the starter?

One email, with the link to the kit.

Sent. Check your inbox for the link. If it is not there in a few minutes, look in spam.

What it does not do

Whisper can mishear names and jargon, so read the captions before you post anything. This is not a full editor: it does not cut dead air, it does not crop to the speaker, and it does not tell speakers apart. A wide screen recording sits small in the vertical frame, so on-screen text can be hard to read. If other people are on the recording, get their yes before you post a clip.

Our code is MIT licensed. Remotion, which draws the captions, has its own licence: free for individuals and teams of up to 3 people, and a company of 4 or more needs a Company License. You are responsible for checking that you qualify, and Remotion may change its terms in a future version. Remotion licence ยท Company License

Not ready to code?

Descript, where you edit the transcript like a document, or CapCut, with automatic word-by-word captions, will get you most of the way with no setup. This kit earns its place once you are cutting the same kind of clip again and again.

This starter was built by The Quiet Scribe team. For help installing it or learning it, contact The Quiet Ai, the mind behind The Quiet Scribe.

Visit thequietai.com

Free community, if you would rather learn alongside others: Wellness & AI, on Skool.